The present disclosure relates to optical character recognition (OCR). Computing devices implement OCR to recognize text characters in an image. For example, when an optical scanner scans a paper document and generates a raster image of the document, an OCR tool can search the raster image for a region that resembles text and then identify characters of text in that region. A conventional OCR process identifies a predominant orientation of text on a page and then recognizes text characters having the predominant orientation on the page. For example, a conventional OCR process can detect that text is oriented at ninety degrees on a page and then recognize text characters having that orientation. The predominant orientation is determined by attempting character recognition for a sample region of a page at multiple orientations (e.g., 0°, 90°, 180°, and 270°) and choosing the page orientation based on the character recognition results for the sample region. In some cases, a conventional OCR process identifies the predominant orientation by calculating an aspect ratio of the input image, by calculating a histogram of the input image and/or by identifying a text baseline in the image.